There is no "rest of the data."
Perhaps it's easiest to demonstrate with an example. Take this file:
http://www.lindberg.no/hires/mqa-2018/2L-110_04_stereo.mqa.flac
First, check the MQA metadata:
Code:
$ mqascan -1 2L-110_04_stereo.mqa.flac
00000000: MQA signature at bit 8
00000000: [5] datasync
magic 36: 0x11319207d
stream_pos_flag 1: 0
pad 1: 1
orig_rate 5: 0x03 [352.8 kHz]
src_rate 5: 0x00 [44.1 kHz]
render_filter 5: 6
unknown_1 2: 0
render_bitdepth 2: 1 [18 bits]
unknown_2 4: 0x0
auth_info 4: 0x0
auth_level 4: 0x9
item_count 7: 2
size 8: 0x14
size 8: 0x0b
type 8: 0x00
type 8: 0x01
[type 0]
stage2_dither 2: 2
gain_index 4: 0
unknown_5 7: 15
unknown_6 7: 127
[type 1]
unknown_7 6: 25
unknown_8 2: 1
unknown_9 1: 0
unknown_10 2: 1
checksum 4: 0xc
This tells us that the original file had a sample rate of 352.8 kHz, the MQA file is 44.1 kHz, and that the "renderer" output should have a resolution of 18 bits. It also instructs the decoder to turn on the blue light.
Next, decode and "render" the file to 352.8 kHz:
Code:
$ mqadec 2L-110_04_stereo.mqa.flac dec.wav
mqaAuthored/88200
$ mqarender -r 352800 dec.wav rend.wav
$ soxi dec.wav rend.wav
Input File : 'dec.wav'
Channels : 2
Sample Rate : 88200
Precision : 24-bit
Duration : 00:05:42.88 = 30242016 samples ~ 25716 CDDA sectors
File Size : 181M
Bit Rate : 4.23M
Sample Encoding: 24-bit Signed Integer PCM
Input File : 'rend.wav'
Channels : 2
Sample Rate : 352800
Precision : 24-bit
Duration : 00:05:42.88 = 120968064 samples ~ 25716 CDDA sectors
File Size : 726M
Bit Rate : 16.9M
Sample Encoding: 24-bit Signed Integer PCM
The decoder outputs an 88.2 kHz file, and the "renderer" upsamples this to the requested 352.8 kHz.
Take a closer look at that final output:
Code:
$ sox rend.wav -n stats
Overall Left Right
DC offset -0.000000 0.000000 -0.000000
Min level -0.892944 -0.849609 -0.892944
Max level 0.931061 0.909576 0.931061
Pk lev dB -0.62 -0.82 -0.62
RMS lev dB -21.89 -21.80 -21.98
RMS Pk dB -9.90 -10.38 -9.90
RMS Tr dB -69.40 -69.40 -69.00
Crest factor - 11.19 11.69
Flat factor 2.92 4.44 0.00
Pk count 2.50 3 2
Bit-depth 16/16 16/16 16/16
Num samples 121M
Length s 342.880
Scale max 1.000000
Window s 0.050
Well, ain't that funny. There are only 16 bits actually in use here.
As we know that "rendering" is just upsampling with a strange filter that might affect the passband, taking that final version (that would have been sent to the DAC chip) and downsampling it to 88.2 kHz using a proper filter will preserve all real information that the MQA process provided. To preserve the dynamic range (dither/quantisation noise density), we give it 2 more bits of resolution, one for each halving of the sample rate.
Code:
$ sox dec.wav alt.flac rate -u 88200 dither -p 18
Finally, we can compare the size of the MQA file to the equivalent plain FLAC:
Code:
$ ls -l *.flac
-rw-r--r-- 1 mru users 57976825 Mar 6 22:57 2L-110_04_stereo.mqa.flac
-rw-r--r-- 1 mru users 54312047 Mar 6 23:19 alt.flac
The FLAC file is only 94% of the MQA file size. In conclusion, MQA is useless for saving bandwidth.